Meta has stepped up its challenge to the leading artificial intelligence companies with the launch of Muse Glimmer, a new open-weight AI model designed to run directly on personal computers.
The release marks a renewed push by Meta CEO Mark Zuckerberg to make advanced AI technology more accessible to developers and users rather than keeping powerful models exclusively behind corporate-controlled systems.
Muse Glimmer is a 30-billion-parameter model built for agentic tasks, meaning it is designed to do more than simply generate answers. It can be used for tasks such as coding, research, planning and other multi-step activities that require an AI system to work through a problem and complete actions.
One of the model’s biggest selling points is its ability to operate locally. Meta says Muse Glimmer can run on a single graphics processing unit, allowing developers to use the model on consumer hardware rather than depending entirely on expensive cloud infrastructure.
That could make a difference for developers and businesses that want greater control over their AI systems. Running models locally can reduce reliance on cloud services, potentially lower costs and give users greater control over data and how an AI model is customised.
Muse Glimmer is also designed around local AI agents. These systems can perform tasks on behalf of users instead of simply responding to individual prompts. The approach is increasingly becoming a major focus of the AI industry, with companies looking at AI agents that can handle longer workflows with less human intervention.
Meta developed Glimmer using a technique known as distillation, drawing capabilities from its more powerful Muse Spark model. This allows a smaller system to retain useful capabilities while being efficient enough to run on consumer hardware.
The launch is part of a broader change in Meta’s AI strategy. Earlier this year, the company introduced Muse Spark as a more powerful model, but Glimmer represents a return to Meta’s open-weight approach. The company has also said it plans to make a more advanced version, Muse Spark 1.2, available with its weights.
Open-weight models give developers access to the underlying model parameters, allowing them to run, modify and customise AI systems within the terms of their licences. This differs from closed AI models, where users generally interact with the system through a company-controlled service or API.
Zuckerberg used the launch to make a wider argument about how AI should develop. In a lengthy essay titled The Future Is for Everyone, he argued that increasingly powerful AI should not be controlled by a small number of companies or governments.
He said broader access could give individuals more control over AI and help developers create personalised systems for education, work, entrepreneurship and other areas of daily life. His vision centres on what he calls personal superintelligence, AI systems that can be tailored to individual users rather than designed only for large organisations.
Zuckerberg also framed open-weight AI as an issue of global competition. He argued that the United States should avoid policies that place domestic AI developers at a disadvantage compared with Chinese companies developing and distributing open models.
The argument comes as the AI race between the US and China becomes increasingly competitive. Chinese companies have gained attention for producing capable models that can be offered at relatively low cost, putting pressure on US technology companies to improve both performance and accessibility.
Meta’s latest move therefore has both a technology and geopolitical dimension. The company wants developers to adopt its AI models while also arguing that a more open ecosystem can help the United States maintain its position in artificial intelligence.
The strategy is not without risks. Open-weight AI models can be customised in ways that are harder for their creators to control. Critics have raised concerns about misuse, cybersecurity and the possibility that increasingly capable models could be adapted for harmful purposes.
Meta’s approach contrasts with the more controlled strategies of companies such as OpenAI and Anthropic, which have generally kept their most powerful models behind managed services and safety systems.
The company is also investing heavily in the infrastructure needed to support its broader AI ambitions. Alongside the Muse Glimmer announcement, Meta said it would establish a $1 billion fund for communities affected by the expansion of its data-centre network.
For Meta, the challenge is not simply producing another AI model. The company is trying to establish a position in a market dominated by intense competition from OpenAI, Google, Anthropic and rapidly advancing Chinese AI developers.
Muse Glimmer gives Meta another route into that competition. Instead of focusing solely on larger models that require expensive cloud infrastructure, the company is betting on AI that can operate closer to the user.
For developers, the attraction is flexibility. For consumers, the potential benefits include lower dependence on cloud services and greater control over personal data. For Meta, wider adoption could strengthen its influence over the next generation of AI development.